ThinkAloud

Landing Page

Inspiration

Interview prep is awkward.

You either talk to yourself in your room, scroll through random question banks, or ask a friend to pretend to care while you explain your projects for the fifth time.

We wanted to build something better — a tool that feels like practicing with a real tutor who actually listens, responds, and challenges you. ThinkAloud was inspired by the idea of making interview preparation interactive, voice-based, and personalized — because real interviews aren’t multiple choice, they’re conversations.

What It Does

ThinkAloud is an AI-powered interview preparation platform with multiple expert modes.

Users can:

Upload their resume and receive tailored interview questions based on their real experience
Practice explaining past coding projects conversationally
Engage in live voice-based mock interviews
Receive AI-generated feedback and session summaries

Instead of static question banks, ThinkAloud simulates a dynamic interview loop. It listens, thinks, responds, and adapts in real time — just like an actual interviewer (but nicer).

How We Built It

Frontend

React + TypeScript
Vite for fast iteration and development

Backend

Node.js + Python services
WebSocket streaming for real-time communication

AI & Infrastructure

Gemini API for generating contextual interview questions and feedback
ElevenLabs API for real-time speech-to-text and text-to-speech
MongoDB for storing user sessions and interaction history

The system streams audio input to the backend, converts speech to text in real time, generates AI responses dynamically, and streams voice output back to the user — creating a seamless conversational loop.

Challenges We Ran Into

Integrating multiple AI APIs while keeping latency low
Handling real-time WebSocket streaming for audio input/output
Managing session state without triggering audio feedback loops
Synchronizing UI updates with streaming AI responses

Real-time voice systems are significantly more complex than traditional request-response apps. Coordinating speech recognition, LLM generation, and audio playback simultaneously required careful state management and debugging.

Accomplishments That We're Proud Of

Building a fully functional live voice interview loop
Making resume-aware question generation contextually accurate
Designing a modular “Expert Mode” system that can scale
Delivering a polished MVP within hackathon time constraints

We didn’t just build a chatbot — we built a multi-modal, interactive AI interview simulator.

What We Learned

How to architect real-time AI systems with WebSockets
How to integrate and orchestrate multiple AI services effectively
The importance of UX in voice-first applications (latency matters a lot)
How strong collaboration under time pressure leads to clean system design

We also learned that when multiple AI services are involved, debugging becomes… an adventure.

What’s Next for ThinkAloud

Expanding Expert Modes (Behavioral Coach, Coding Interviewer, System Design Mentor)
Adding user authentication and long-term performance tracking
Building structured scoring and analytics dashboards
Optimizing streaming efficiency and latency
Deploying scalable cloud infrastructure

Our long-term vision is to create a personalized AI interview coach that evolves with the user — adapting to their career path, skill level, and goals over time.